fix: gate dotenv loading + improve api logging

src/api/logging/ApiInferenceLogger.ts

@@ -27,6 +27,13 @@ export interface ApiInferenceHandle {
 	error: (errorPayload: unknown) => void
 }
 
+function extractModelFromPayload(payload: unknown): string | undefined {
+	if (!payload || typeof payload !== "object") return undefined
+	const rec = payload as Record<string, unknown>
+	const model = rec["model"]
+	return typeof model === "string" && model.trim().length > 0 ? model : undefined
+}
+
 /**
  * Configuration for payload size limiting to avoid freezing the Output Channel.
  */
@@ -245,11 +252,13 @@ class ApiInferenceLoggerSingleton {
 		const requestId = context.requestId ?? generateRequestId()
 		const startTime = Date.now()
 		const startTimestamp = new Date().toISOString()
+		const resolvedModel = context.model ?? extractModelFromPayload(requestPayload)
 
 		// Log the request
 		if (this.isEnabled()) {
 			this.logRequest({
 				...context,
+				...(resolvedModel ? { model: resolvedModel } : {}),
 				requestId,
 				timestamp: startTimestamp,
 				payload: requestPayload,
@@ -262,6 +271,7 @@ class ApiInferenceLoggerSingleton {
 					const endTime = Date.now()
 					this.logResponse({
 						...context,
+						...(resolvedModel ? { model: resolvedModel } : {}),
 						requestId,
 						timestamp: new Date().toISOString(),
 						durationMs: endTime - startTime,
@@ -274,6 +284,7 @@ class ApiInferenceLoggerSingleton {
 					const endTime = Date.now()
 					this.logError({
 						...context,
+						...(resolvedModel ? { model: resolvedModel } : {}),
 						requestId,
 						timestamp: new Date().toISOString(),
 						durationMs: endTime - startTime,

src/api/logging/__tests__/ApiInferenceLogger.spec.ts

@@ -40,10 +40,7 @@ describe("ApiInferenceLogger", () => {
 			ApiInferenceLogger.start({ provider: "OpenAI", operation: "createMessage" }, { model: "gpt-4" })
 
 			expect(mockSink).toHaveBeenCalledTimes(1)
-			expect(mockSink).toHaveBeenCalledWith(
-				"[API][request][OpenAI][unknown]",
-				expect.objectContaining({ model: "gpt-4" }),
-			)
+			expect(mockSink).toHaveBeenCalledWith("[API][request][OpenAI][gpt-4]", { model: "gpt-4" })
 		})
 
 		it("should use context.model in the request label", () => {
@@ -63,6 +60,21 @@ describe("ApiInferenceLogger", () => {
 				expect.objectContaining({ test: "data" }),
 			)
 		})
+
+		it("should fall back to payload.model when context.model is missing", () => {
+			ApiInferenceLogger.start(
+				{
+					provider: "OpenAI",
+					operation: "createMessage",
+				},
+				{ model: "gpt-4", foo: "bar" },
+			)
+
+			expect(mockSink).toHaveBeenCalledWith(
+				"[API][request][OpenAI][gpt-4]",
+				expect.objectContaining({ model: "gpt-4", foo: "bar" }),
+			)
+		})
 	})
 
 	describe("success", () => {
@@ -202,13 +214,10 @@ describe("ApiInferenceLogger", () => {
 				{ model: "gpt-4", messages: [{ role: "user", content: "Hello" }] },
 			)
 
-			expect(mockSink).toHaveBeenCalledWith(
-				"[API][request][test][unknown]",
-				expect.objectContaining({
-					model: "gpt-4",
-					messages: [{ role: "user", content: "Hello" }],
-				}),
-			)
+			expect(mockSink).toHaveBeenCalledWith("[API][request][test][gpt-4]", {
+				model: "gpt-4",
+				messages: [{ role: "user", content: "Hello" }],
+			})
 		})
 	})
 

src/api/logging/logging-fetch.ts

@@ -741,10 +741,11 @@ export function createLoggingFetch(options: LoggingFetchOptions, baseFetch: type
 
 			// Tee the stream so the consumer can read one branch while we buffer the other.
 			const [consumerStream, logStream] = response.body.tee()
+			// Note: returning a new Response means some Response properties (e.g. `url`) are not preserved.
 			const responseForConsumer = new Response(consumerStream, {
 				status: response.status,
 				statusText: response.statusText,
-				headers: response.headers,
+				headers: new Headers(response.headers),
 			})
 			void logSseResponse(logStream, options.provider, model, startedAt, maxBytes)
 			return responseForConsumer

src/api/providers/openai-native.ts

@@ -181,14 +181,8 @@ export class OpenAiNativeHandler extends BaseProvider implements SingleCompletio
 			metadata,
 		)
 
-		let lastUsage: any
-
 		// Make the request (pass systemPrompt and messages for potential retry)
 		for await (const chunk of this.executeRequest(requestBody, model, metadata, systemPrompt, messages)) {
-			if (chunk.type === "usage") {
-				lastUsage = chunk
-			}
-
 			yield chunk
 		}
 	}

src/extension.ts

@@ -1,18 +1,27 @@
 import * as vscode from "vscode"
 import * as dotenvx from "@dotenvx/dotenvx"
 import * as path from "path"
+import * as fs from "fs"
 
-// Load environment variables from .env and .env.local files
+// Load environment variables from local development files if present.
+// In packaged production builds these files typically won't exist.
 try {
 	// Specify paths to .env and .env.local files in the project root directory
 	const envPath = path.join(__dirname, "..", ".env")
 	const envLocalPath = path.join(__dirname, "..", ".env.local")
 	// Load .env first, then .env.local (so .env.local can override)
-	dotenvx.config({ path: envPath })
-	dotenvx.config({ path: envLocalPath, override: true })
+	if (fs.existsSync(envPath)) {
+		dotenvx.config({ path: envPath })
+	}
+	if (fs.existsSync(envLocalPath)) {
+		dotenvx.config({ path: envLocalPath, override: true })
+	}
 } catch (e) {
-	// Silently handle environment loading errors
-	console.warn("Failed to load environment variables:", e)
+	// Don't let dotenv errors break extension activation.
+	// Avoid noisy logs for normal users.
+	if (process.env.NODE_ENV === "development") {
+		console.warn("Failed to load environment variables:", e)
+	}
 }
 
 import type { CloudUserInfo, AuthState } from "@roo-code/types"